Home, Optimizer, Benchmarks, Server Systems, Systems Architecture, Processors, Storage,
  Storage Overview, System View of Storage, SQL Server View of Storage, File Layout,

PCI-ESASFCHDDSSD Technology RAID ControllersDirect-Attach
  SAN,  Dell MD3200,  CLARiiON AX4CX4VNXV-Max,  HP P2000EVAP9000/VSP,  Hitachi AMS
  SSD products: SATA SSDsPCI-E SSDsFusion iOother SSD

Hard Drives

Seagate, Hitachi and Toshiba are the primary enterprise hard disk manufacturers remaining. Fujitsu used to make enterprise hard drives, but transferred all HD business to Toshiba. The Hitachi hard disk business was acquired from IBM some years back, now under Hitachi Global Storage Technologies, not HDS.

Tiered Storage

In the last few years, it has become popular among expensive storage system vendors to talk about tiered storage. The tiers range from expensive high-performance storage to low cost high-capacity 7200 RPM disks. This started with 15K - 10K - 7200RPM, but has evolved to SSD - 15K - 7200RPM.

Seagate HDD

The Seagate Hard Disk Drives product lines consist of 15K, 10K and 7200RPM models in both 2.5in and 3.5in form factors, which HP call SFF and LFF respectively. The Cheetah brand is for 3.5in 15K and 10K (Cheetah NS) models. The Savvio brand comprises 2.5in 15K and 10K models. The Constellation brand consists of 7200RPM drives in both 3.5in and 2.5in form factors.

Below is bare disk drive pricing from NewEgg, (2010-10-18).

Form Factor
 2.5in3.5in
RPMModelGB$ModelGB$
15K Savvio
15K.2
73GB $190 Cheetah
15K.7
300GB $270
146GB $240450GB $410
  600GB $510
10K Savvio
10K.4
450GB $450Cheetah NS450GB $380
600GB $460600GB $460
7200 Constellation? Constellation500GB $92
? 1TB $150
500GB $1802TB $310

The 7200RPM Constellation lines are "enterprise grade" drives based on desktop and notebook models. Presumably the drives meet more stringent quality control specifications. It is unclear if there are any actual component differences. The more powerful magnet used in the 10K drives would certainly be welcome. Compare with the consumer-desktop Barracuda 3.5in 7200RPM 2TB at $200, and $100 for the Barracuda 5900RPM at $100. The Savvio 10K.3 300GB is $240 and the 146GB is $150.

Seagate Cheetah 10K and 15K 3.5in HDs

The latest high performance 3.5in drive is the Cheetah 15K.7 has 300, 450 and 600GB models. The Cheetah 15K.7 comes in 300, 450 and 600GB models, with 2, 3, and 4 platters.

Development on the 3.5in Cheetah 10K lines stopped with the seventh generation, which was contemporary with the 15K.4 model(?), a 2004/5 product. Then recent a Cheetah NS10K.2 appears (perhaps there was a prior NS10K.1 that I missed), with capacities 450 and 600GB on 3 and 4 platters. I am thinking that this is simply the Cheetah 15K.4 adapted to run at 10K.

If it were a true 10K product, then the platter would be larger and the media density would be higher. Historically, the 10K disks were twice the 15K disk capacity for the same number of platters. And the 3.5in disks were typically twice the capacity (or more) of their 2.5in counter-parts. Areal density is 225Gb/in2 versus 343 on the Savvio 10K.4.

Savvio

The Savvio line started with 10K models, and later 15K models were introduced. Seagate is currently on the 4th generation Savvio 10K and 2nd generation Savvio 15K. The Savvio 10K.4 comes in 450 and 600GB models, both 3 platters. The Savvio 10K.3 with 146G and 300GB models are still active? with 1 and 2 platters. The Savvio 15K.2 comes in 73 and 146GB models with 1 and 2 platters. The areal density of the 600GB 10K.4 is 343Gb/in2 versus 257Gb/in2 for the 450GB 10K.4, a rather unusual aspect for drives of the same model and generation.

Constellation

The 7200RPM Constellation line is offered in 2.5in at 500GB and 160GB. The 3.5in models run up to 2TB.

Below are the Seagate Max sustained sequential transfer and seek times.

Form FactorMax CapacitySequential BWAvg. Seek
RPM2.5in3.5in2.5in3.5in2.5in3.5in
7200500GB2TB?-10060-1508.0/8.48.3/9.3
10K600GB600GB 72-135 82-1503.8/4.43.8-4.4
15K146GB600GB105-152122-2052.9/3.33.4/3.9

 Cheetah 15K.6Cheetah 15K.7Savvio 15K.2
  Avg. Read Seek3.4ms3.4ms2.9ms
  Avg. Write Seek3.9ms3.9ms3.3ms
  Sequential Max171MB/s204MB/s160MB/s
  Sequential Min112MB/s122MB/s122MB/s

There are 640GB and 1TB 2.5in SATA drives (not 7200RPM?) but these are currently not available in the Enterprise line. In general, I recommend the 15K and 10K drives for databases, despite the higher cost per disk. If the price is an issue, I suggest not buying grossly over-priced storage systems with pathetic performance, and just buy quality components for direct-attach storage.

Toshiba HDD

The former Fujitsu, now Toshiba enterprise hard disk comprise 3.5in and 2.5in form factors, 10K and 15K lines. The 3.5in form factor tops out at 300GB for both 10K and 15K, indicating this form factor is probably on end of life. The 2.5in form factor has 10K models at 300, 450 and 600GB in the MBF2xxx RC series. The 2.5in 15K MBE2xxx RC series. are in 73 and 146GB models.

Hitachi HDD

Hitachi has 3.5in 15K hard drives to 600GB, 2.5in 15K drives to 146GB, 2.5in 10K drives to 600GB and 7200RPM 3.5in drives to 2TB.

Hard Drives - Older

The table below shows the specifications for the recent Seagate 3.5in (LFF) and 2.5in (SFF) 15K drives. The 2.5in Savvio drive has lower average seek time. The rotational latency for 15K drives is 2.0ms. The transfer time for an 8KB block ranges from 0.04ms at 204MB/s to 0.065ms at 122MB/s. The average access time for 8K IOP randomly distributed over the entire disk is then 5.45ms for the 3.5in disk and 4.95ms for the 2.5in disk. It should also be considered that the 3.5in Cheetah 15K.7 has media density of 150GB per platter versus 73GB for the 2.5in Savvio 15K.2. If the 3.5in disk were only populated to 50% capacity, the average seek latency would probably be comparable with the 2.5in disk.

The sequential transfer rates assume no errors and no relocated logical blocks. On the enterprise class disk drives, this is effectively achieved. On the high-capacity 7200RPM drives, the ability to sustain the perfect transfer rates is highly problematic, and the data sheet may not specify the design transfer rate.

The chart below shows IOMeter results for a single 10K over a range of data space utilization and queue depth demonstrating the short-stroke effect on IOPS (vertical axis).

10K IOPS Q

The charts below show latency on the vertical scale in ms for a range of data utilizations and queue depth.

10K lat Q

10K lat Q

The figures below show both the short-effect and queue depth impact on IOPS and access latency for a 15K disk.

10K IOPS Q
IOPS versus queue depth for various disk space utilizations

10K lat Q
Latency versus queue depth for various disk space utilizations

more IO Cost Structure

There is no point to having the big capacity SATA disks in the main storage system. We said early that the short-stroke effect was key. This meant we will have much more space than needed on the set of 15K drives. The SATA drives are good for allowing dev and QA to work with the full database. There are too many developers who cannot understand why a query works fine on a tiny 10MB dev database, but not the 10TB production database.

Short Stroke Effect

The short-stroke effect is absolutely essential for transaction processing systems with tight mandatory limits on responsiveness. The short-stroke effect lowers latency and improves random IO performance. Most importantly, the short-stroke effect keeps latency low during heavy IO surges when active data is kept in a very narrow band of the disk. On a fully populated disk where full strokes are required, latency can jump to several hundred milli-sec during heavy IO surges.

One of the fundamental arguments made by SAN storage vendors is that by consolidating storage, it is possible to achieve high storage utilization, i.e., guaranteeing the full-stroke criteria. A heavy IO surge is very likely to cause transaction processing volume to collapse. To benefit from the short-stroke effect, it is necessary to restrict the active data to a very narrow range. The remaining disk capacity can still be used for data not in use during busy hours. This means having far more capacity than the active database, which in turn implies that it is essential to keep amortized cost per disk low, i.e., forgoing frills.

Single Hard Disk Drive Performance (2009)

A very brief discussion of disk drive performance characteristics is warranted. The difference between random and sequential IO is well known if not fully appreciated. Recent generation 10,000rpm (10K) disk drives can sustain sequential transfer rates of 80MB/sec on the outer tracks and 40MB/sec on the inner tracks. For 15K drives, 125MB/s on the outer and 70MB/s on the inner tracks. Random IO performance is constrained by disk rotational speed and seek-time.

The disk media on a 10K drive completes a full rotation in 6ms, for an average rotational latency contribution of 3ms, and 2ms for a 15K drive. Current generation 10K drives have an average seek time of 4.6ms for reads and 5.2ms for writes. A 15K drive might have average seek times of 3.5ms for reads and 4.2ms for writes. For a small 8KB access, the transfer time is approximately 0.1ms. Other contributions are negligible, so the total latency for random reads is about 8ms on 10K and 5.6ms on 15K drives. This leads to the frequently quoted small block random IO performance of 125 IOPS for 10K and 175 IOPS for 15K drives respectively.

Implied in the above calculations are that disk IO is issued serially one at a time and data is distributed across the entire disk. When data is restricted to a narrow section, then the average seek time is lower, the short stroke effect. When multiple disk IO requests are issued simultaneously, the disk re-orders the IO sequence, command queuing, for higher throughput at the expense of longer latency. If data is distributed over the entire disk, high-queue depth operation quickly leads to very high latency, exceeding 20ms at queue depth 4 and over 100ms at queue depth 32 per disk.

This is the underlying cause and effect of the maximum queue depth 2 recommendation. A very important matter rarely discussed is the short-stroke effect combined with high queue depth operation. When only a tiny fraction of the disk is used for the active database, preferably 5% or less, latency increases much more slowly with queue depth, less than 20ms at queue depth 8 and less than 40ms at queue depth 16 and random IO performance can exceed 400 IOPS per disk. This is characteristic is the key in maintaining strong transaction processing resiliency. This is also counter the arguments made for SAN in allowing high disk space utilization.

Disk Images

Below are images for (older models) of the Seagate 3.5in drives at 7200, 10K and 15K RPM. Notice that the 7200RPM platter fills the entire case. The platter in the 10K drive is visibly smaller, and the 15K platter is even smaller.

Seagate 7200
Seagate 7200RPM 3.5in drive

Seagate 10K Seagate 15K
Seagate 10K 3.5in drive and 15K 3.5in drive

Additional images for the Seagate 7200RPM and 15K 3.5in drives.

Seagate Barracuda Seagate Cheetah 15K

        Seagate 7200RPM 3.5in drive                         Seagate 15K 3.5in drive

Images for the Seagate 10K and 15K 2.5in drives.

Seagate Savvio Seagate Savvio 15K

        Seagate Savvio 10K 2.5in drive         Seagate Cheetah 15K 2.5in drive

In the images Seagate shows for the most recent Savvio drive, the 15K now seems to employ a larger platter.

Hitachi drive images

Below are the 3.5in Hitachi drives at 15K and 7200RPM. With consideration for the different scale, the 15K disk platter is much smaller than the 3.5in case (which is actually 4in) while the 7200RPM platter fills the case.

Below are the 2.5in Hitachi drives at 15K and 10K RPM. Notice that the 15K platter is smaller than the 2.5in case (actually 2.76in) while the 10K platter fills the case.

This is an old reference fron NT 4.0 days, but defines a number of terms and has useful diagrams NT Server and Disk Subsystem Performance

LSI 1078 LSI 1078